On 3 September 2026 OpenAI released GPT-6 Astra. It says the model is strongest on computer use, browsing, software, science, and professional work, and that it is the first to reach the Critical cybersecurity level in OpenAI’s Preparedness Framework.
Rollout started with a limited set of organizations, then ChatGPT paid plans, the API as gpt-6-astra, Azure, and Bedrock. Enterprise access is off by default. A safety overview was published the same day.
Saturated benches are vendor scores. Critical capability is a reason to tighten access, not to skip a human check on anything that can bind or harm.
OriginalOpenAIopenai.comRead the full story